1

Comparison of Routing Metrics for Static Multi-Hop Wireless Networks

Richard Draves, Jitendra Padhye and Brian Zill

Microsoft Research

2

Multi-hop Wireless Networks

Static Mobile

Motivating scenario

Community wireless networks (“Mesh Networks”)

Battlefield networks

Key challengeImproving network

capacity

Handling mobility, node failures, limited

power.

3

Routing in Multi-hop Wireless Networks

• Mobile networks: – Minimum-hop routing (“shortest path”)– DSR, AODV, TORA ….

• Static networks: – Minimum-hop routing tends to choose long, lossy

wireless links – Taking more hops on better-quality links can improve

throughput [De Couto et. al., HOTNETS 2003]

4

Link-quality Based Routing

• Metrics to measure wireless link quality:– Signal-to-Noise ratio– Packet loss rate– Round trip time– Bandwidth– …

Our paper: experimental comparison of performance of three metrics in a 23 node, indoor testbed.

5

Contributions of our paper

• Design and implementation of a routing protocol that incorporates notion of link quality– Link Quality Source Routing (LQSR)– Operates at layer “2.5”

• Detailed, “side-by-side” experimental comparison of three link quality metrics: – Per-hop Round Tip Time (RTT) [Adya et al 2004]– Per-hop Packet Pair (PktPair)– Expected Transmissions (ETX) [De Couto et al 2003]

6

Summary of Results

• ETX provides best performance

• Performance of RTT and PktPair suffers due to self-interference

• PktPair suffers from self-interference only on multi-hop paths

7

Outline of the rest of the talk

• LQSR architecture (brief)

• Description of three link quality metrics

• Experimental results

• Conclusion

8

LQSR Architecture

• Source-routed, link-state protocol– Derived from DSR

• Each node measures the quality of links to its neighbors

• This information propagates throughout the mesh

• Source selects route with best cumulative metric

• Packets are source-routed using this route

9

Link Quality Metrics

1. Per-hop Round Trip Time (RTT)

2. Per-hop Packet-Pair (PktPair)

3. Expected transmissions (ETX)

4. Minimum-hop routing (HOP)• Binary link quality

10

Metric 1: Per-hop RTT

• Node periodically pings each of its neighbors– Unicast probe/probe-reply pair

• RTT samples are averaged using TCP-like low-pass filter

• Path with least sum of RTTs is selected

11

Metric 1: Per-hop RTT

• Advantages– Easy to implement– Accounts for link load and bandwidth– Also accounts for link loss rate

• 802.11 retransmits lost packets up to 7 times• Lossy links will have higher RTT

• Disadvantages– Expensive – Self-interference due to queuing

12

Metric 2: Per-hop Packet-Pair

• Node periodically sends two back-to-back probes to each neighbor– First probe is small, second is large

• Neighbor measures delay between the arrival of the two probes; reports back to the sender

• Sender averages delay samples using low-pass filter

• Path with least sum of delays is selected

13

Metric 2: Per-hop Packet-Pair

• Advantages– Self-interference due to queuing is not a problem– Implicitly takes load, bandwidth and loss rate into

account

• Disadvantages– More expensive than RTT

14

Metric 3: Expected Transmissions

• Estimate number of times a packet has to be retransmitted on each hop

• Each node periodically broadcasts a probe– 802.11 does not retransmit broadcast packets

• Probe carries information about probes received from neighbors

• Node can calculate loss rate on forward (Pf) and reverse (Pr) link to each neighbor

• Select the path with least total ETX

)P1(*)P1(

1

rf ETX

15

Metric 3: Expected Transmissions

• Advantages– Low overhead– Explicitly takes loss rate into account

• Disadvantages– Loss rate of broadcast probe packets is not the same

as loss rate of data packets• Probe packets are smaller than data packets• Broadcast packets are sent at lower data rate

– Does not take data rate or link load into account

16

Mesh Testbed

Approx. 61 m

Appro

x.

32

m

23 Laptops running Windows XP. 802.11a cards: mix of Proxim and Netgear.

Diameter: 6-7 hops.

17

Link bandwidths in the testbed

0

5

10

15

20

25

30

0 5 10 15 20 25 30Higher Bandwidth (Mbps)

Lo

wer

Ban

dw

dit

h (

Mb

ps)

• Cards use Autorate •Total node pairs: 23x22/2 = 253

• 90 pairs have non-zero bandwidth in both directions.

Bandwidths vary significantly; lot of asymmetry.

18

Experiments

1. Bulk-transfer TCP Flows

2. Impact of mobility

19

Experiment 1

• 3-Minute TCP transfer between each node pair– 23 x 22 = 506 pairs– 1 transfer at a time– Long transfers essential for consistent results

• For each transfer, record: – Throughput– Number of paths

• Path may change during transfer

– Average path length• Weighted by fraction of packets along each path

20

Median Throughput

0

200

400

600

800

1000

1200

1400

1600

HOP ETX RTT PktPair

Med

ian

Th

rou

gh

pu

t (K

bp

s)

ETX performs best. RTT performs worst.

21

Why does ETX perform well?

0

0.2

0.4

0.6

0.8

1

0 2000 4000 6000 8000 10000

Throughput (Kbps)

Cu

mu

lati

ve

Fra

cti

on

ETX

HOP

ETX performs better by avoiding low-throughput paths.

22

0

1

2

3

4

5

6

7

8

0 1 2 3 4 5 6 7 8

Path Length with ETX

Pat

h L

eng

th w

ith

HO

P

Impact on Path Lengths

Path length is generally higher under ETX.

23

Why does RTT perform so poorly?

RTT suffers heavily from self-interference

Median Number of Paths

0

5

10

15

20

25

HOP ETX RTT PktPair

Nu

mb

er o

f P

ath

s

24

What ails PktPair?

PktPair

0

2000

4000

6000

8000

10000

12000

0 1 2 3 4 5 6 7 8

Average Pathlength (Hops)

Th

rou

gh

pu

t (K

bp

s)

ETX

0

2000

4000

6000

8000

10000

12000

0 1 2 3 4 5 6 7 8

Average Path Length (Hops)

Th

rou

gh

pu

t (K

bp

s)

RTT

0

2000

4000

6000

8000

10000

12000

0 1 2 3 4 5 6 7 8

Average Path Length (Hops)

Th

rou

gh

pu

t (K

bp

s)

PktPair suffers from self-interference only on multi-hop paths.

25

Summary

• ETX performs well despite ignoring link bandwidth

• Self-interference is the main reason behind poor performance of RTT and PktPair.

Similar results for multiple simultaneous flows.

26

Experiment 2

• Walk slowly around network periphery for 15 minutes with a laptop

• Mobile laptop is the sender, a corner node is receiver

• Repeated 1-minute TCP transfers

27

Testbed Layout

Approx. 61 m

Appro

x.

32

m

28

0

100

200

300

400

500

600

HOP ETX

Metric

Me

dia

n T

CP

Th

rou

gh

pu

t (K

bp

s)

Shortest path routing is best in mobile scenarios?

29

Conclusions

• ETX metric performs best in static scenarios

• RTT performs worst

• PacketPair suffers from self-interference on multi-hop paths

• Shortest path routing seems to perform best in mobile scenarios– Metric-based routing does not converge quickly?

30

Ongoing/Future work

• Explicitly take link bandwidth into account

• Support for multiple heterogeneous radios per node– To appear in MOBICOM 2004

• Detailed study of TCP performance in multi-hop networks

• Repeat study in other testbeds

31

For more information

http://research.microsoft.com/mesh/

Source code, binaries, tech reports, …

32

Backup slides

33

LQSR Architecture

• Implemented in a shim layer between Layer 2 and 3.

• The shim layer acts as a virtual Ethernet adapter– Virtual Ethernet addresses– Multiplexes heterogeneous

physical links

• Advantages:– Supports multiple link

technologies– Supports IPv4, IPv6 etc

unmodified– Preserves the link abstraction– Can support any routing

protocol

• Architecture:

• Header Format:

Ethernet 802.11 802.16

Mesh connectivity Layer with LQSR

IPv4 IPv6 IPX

Ethernet MCL

Payload:TCP/IP,

ARP,IPv6…

34

Web transfers

• Simulated Web transfer using Surge

• One node serves as web server

• Six nodes along periphery act as clients

• Results: ETX reduces latency by 20% for hosts that are more than one hop away from server.

35

Static Multi-hop Wireless Networks

• Motivating scenario: – Community wireless networks (“Mesh Networks”)

• Very little node mobility• Energy not a concern

• Main Challenge:– Improve Network capacity

• Minimum-hop count routing is inadequate– Tends to choose long, lossy wireless links [De Couto et.

al., HOTNETS 2003]

36

“Traditional” Multi-hop Wireless Networks

• Envisioned for mobility-intensive scenarios

• Main concerns: – Reduce Power consumption– Robustness in presence of mobility, link failures

• Routing:– Minimum-hop routing (“shortest path”) with various

modifications to address power and mobility concerns– DSR, AODV, TORA ….